What’s the impact of Trump’s latest restrictions on NVIDIA and other chip suppliers?
Portfolio Manager Richard Clode discusses President Trump’s latest semiconductor restrictions, which are advancing the shift towards AI chip deglobalisation.
4 minute read
Key takeaways:
- President Trump’s additional sanctions on the export of AI chips to China has led NVIDIA to take a near-term earnings provision. Existing expectations for the H20 chip’s future prior to this were already short lived.
- NVIDIA and US hyperscalers are lobbying the government to repeal the AI Diffusion Rule arguing that it will impede AI global supremacy ambitions for the US.
- We may be heading closer to a total AI semi US-China decoupling. China now needs its AI accelerators domestically designed and manufactured, which is likely to prove challenging.
NVIDIA recently filed an 8-K form (submitted by listed companies to report significant company changes like a change in accounts or important events) after it received a letter from the US government on 9 April that its H20 lower power AI chips designated for the Chinese market would now be subject to licensing and the list of D5 sanctioned (arms embargoed) countries effective immediately. While that is not a ban in itself, NVIDIA appears to be assuming presumption of denial on those license applications. Hence it is writing off US$5.5 billion of H20 inventory given the only markets for these degraded chips is China.
This has been a long-winded drama that has taken many twists and turns. The market has gone from full acceptance of a ban being the likely outcome (some analysts had already stripped out the H20 chips from future earnings forecasts prior to this news), to more recently, hope that a rumoured Mar-a-Lago dinner between Jensen Huang and Trump had led to a reprieve. This was fuelled by the announcement this week that NVIDIA was committing to building US$500 billion of AI supercomputer infrastructure in the US, with a follow-up tweet from Trump also seeming to feed that narrative. Linked to this, NVIDIA has reportedly been developing a more advanced Blackwell version of the H20 chip, indicating some hope of future China sales.
As ever, in a highly fluid situation with multi-faceted geopolitical considerations, trying to predict these twists and turns is complex.
Key considerations for investors:
1 Are the new restrictions related to the AI Diffusion rule?
The debate now turns to the AI Diffusion Rule that is meant to come into effect on 15 May. There has been a concerted campaign by NVIDIA and all the hyperscalers to push Trump to water down or stop this late Biden administration regulation. The Rule caps the export of essential US AI technology to limit its spread outside the US. The argument being that the rule would impinge America’s ambition for AI global supremacy and be a tailwind to global competitors to fill the void.
Seven Republican senators wrote a letter to Secretary of Commerce Lutnick this week making that point. With US companies aligning with political lobbying for changes, there is a higher chance of it being watered down.
2 AI Capex debate continues
There remains a broader debate on the sustainability of huge AI capital spending even before the macro wobbles. But that has now moved on from AI capex being the most debated issue, to AI capex being a quasi-safe haven and more strategically resilient in a macro slowdown.
3 Broader supply implications
Further Trump semi sanctions were not unexpected. According to supply chain checks, NVIDIA has not placed any new H20 wafer orders to TSMC (the main manufacturer of NVIDIA’s chips) since late 2024. Those orders already placed are being processed at TSMC and sold to Nvidia in H1 2025 and then being scrapped, so the impact on TSMC is likely to be minimal. This new export license decree is not H20 chip specific and applies to all AI accelerators with H20 bandwidth and interconnect specifications, hence it also applies to other chip manufacturers like the AMD MI300 variant being sold in China, and potentially custom projects such as Broadcom’s deal with ByteDance.
AI semiconductor deglobalisation endgame?
It seems we are heading closer to a full AI semiconductor US-China decoupling. AI accelerators in China will now need to be designed and manufactured in China. That remains challenging with more limited expertise, domestic manufacturing capability at 5nm and below, and weaker networking capability.
The void will likely mainly be filled by Huawei’s Ascend chips, with the company having pedigree in advanced processor chip design and advanced networking. However, from an investment point of view, Huawei is not listed and investment in its supply chain can entail higher risk and be problematic.
5nm: 5 nanometer chip refers to an improved generation of silicon semiconductor chips in terms of increased transistor density (i.e. a higher degree of miniaturisation), increased speed and reduced power consumption.
Capex: capital expenditure refers to company spending to acquire or upgrade physical assets such as buildings, machinery, equipment, technology etc. to maintain or improve operations and foster future growth.
Hyperscalers: companies that provide infrastructure for cloud, networking, and internet services at scale. Examples include Google Cloud, Microsoft Azure, Facebook Infrastructure, Alibaba Cloud, and Amazon Web Services.
4 minute read
Key takeaways:
- President Trump’s additional sanctions on the export of AI chips to China has led NVIDIA to take a near-term earnings provision. Existing expectations for the H20 chip’s future prior to this were already short lived.
- NVIDIA and US hyperscalers are lobbying the government to repeal the AI Diffusion Rule arguing that it will impede AI global supremacy ambitions for the US.
- We may be heading closer to a total AI semi US-China decoupling. China now needs its AI accelerators domestically designed and manufactured, which is likely to prove challenging.
